Euroheart Score for the Evaluation of In-Hospital Mortality in Patients Undergoing Percutaneous Coronary Intervention

European Heart Journal (2011) 32, 1398–1408 CLINICAL RESEARCH doi:10.1093/eurheartj/ehr034 Interventional cardiology

EuroHeart score for the evaluation of in-hospital mortality in patients undergoing percutaneous coronary intervention

Maarten de Mulder 1,2, Anselm Gitt 3, Ron van Domburg 2, Matthias Hochadel 3, Ricardo Seabra-Gomes 4, Patrick W. Serruys 2, Sigmund Silber 5, Franz Weidinger 6, William Wijns 7, Uwe Zeymer 3, Christian Hamm 8, and Eric Boersma 2*

1Medisch Centrum Alkmaar, Alkmaar, The Netherlands; 2Thoraxcentrum, Erasmus MC, Rotterdam, The Netherlands; 3Herzzentrum Ludwigshafen, Ludwigshafen, Germany; 4 5 6 7 Instituto do Coracao, Lisbon, Portugal; Cardiology Practice and Hospital, Munich, Germany; Krankenhaus Rudolfstiftung, Vienna, Austria; Onze Lieve Vrouw hospital, Aalst, Downloaded from Belgium; and 8Kerckhoff Klinik, Bad Nauheim, Germany

Received 9 July 2010; revised 30 December 2010; accepted 24 January 2011; online publish-ahead-of-print 22 February 2011

Aims The applicability of currently available risk prediction models for patients undergoing percutaneous coronary inter- http://eurheartj.oxfordjournals.org/ ventions (PCIs) is limited. We aimed to develop a model for the prediction of in-hospital mortality after PCI that is based on contemporary and representative data from a European perspective...... Methods and Our analyses are based on the Euro Heart Survey of PCIs, which contains information on 46 064 consecutive patients results who underwent PCI for different indications in 176 participating European centres during 2005–08. Patients were randomly divided into a training (n ¼ 23 032) and a validation (n ¼ 23 032) set with similar characteristics. In these sets, 339 (1.5%) and 305 (1.3%) patients died during hospitalization, respectively. On the basis of the training set, a logistic model was constructed that related 16 independent patient or lesion characteristics with mortality, including PCI indication, advanced age, haemodynamic instability, multivessel disease, and proximal LAD disease. In by guest on April 2, 2016 both the training and validation data sets, the model had a good performance in terms of discrimination (C-index 0.91 and 0.90, respectively) and calibration (Hosmer–Lemeshow P-value 0.39 and 0.18, respectively)...... Conclusion In-hospital mortality in PCI patients was well predicted by a risk score that contains 16 factors. The score has strong applicability for European practices. ------Keywords Percutaneous coronary intervention † Hospital mortality † Peri-procedural complications † Risk stratiﬁcation † Predictive model

Introduction To identify high-risk patient groups, risk models are developed Since its introduction by the late Andreas Gru¨ntzig in 1979, percu- that relate patient and lesion characteristics to major complications taneous coronary interventions (PCIs) have been applied to the after PCI.2– 8 Especially in situations where it is difﬁcult to select the beneﬁt of millions of patients across the globe. Over the years, most appropriate treatment strategy, they can be of extra value. Risk this procedure has evolved from elective balloon angioplasty in models can then be used to systematically estimate the patient’s risk selected centres to widely available emergency PCI with stent pla- of adverse events. Such estimate might then be used to help the cement. As technology, pharmacology, and operators’ experience physician decide on further patient management, as high-risk with PCI grow, the procedure-associated risks decrease.1 patients might be treated differently than low-risk patients. However, this intervention is still related with mortality, which It is broadly accepted that currently available risk prediction varies between different groups of patients. models for PCI patients have limited applicability, mainly because

* Corresponding author: Department of Cardiology, Clinical Epidemiology Unit, Room Bd-381, PO Box 2040, 3000 CA Rotterdam, The Netherlands. Tel: +31 10 7032307, Email: [email protected] Published on behalf of the European Society of Cardiology. All rights reserved. & The Author 2011. For permissions please email: [email protected] EuroHeart score for the evaluation of in-hospital mortality after PCI 1399 of heavy selection of the patients who form the model- Local investigators were asked to continuously enrol all consecutive development data set. They were either single-centre studies,3,5,8 patients undergoing emergent, urgent, or elective PCI, irrespective of a selected study cohort,2,9 or studies from an era without tech- any other condition. Patients who participated in (randomized) trials niques such as drug-eluting stents; or new anti-platelet medi- or other registries were eligible for inclusion. Investigators who cations.2,4,6,7 These limitations were overcome by Peterson could not warrant enrolment of each and every patient throughout the entire study period were allowed to participate if consecutive et al.,10 who developed a model based on 588 398 procedures patients could be realized from Days 1 to 7 of every calendar from the American NCDR CathPCI Registry database. However, month. We had no system installed to verify whether the principle as this analysis was performed in a geographically different popu- of consecutive patient enrolment was satisfied. 11 lation, its use might be limited for a European population. Data were collected on a broad range of patient characteristics, Additionally, the actual use of risk prediction models in routine including the clinical indication for PCI, cardiovascular risk factors, clinical practice may be an issue. In general, one might expect history of cardiovascular diseases, and co-morbidities. Percutaneous that models that are based on data that are experienced as coronary intervention-related data were collected as well, including ‘close’ will have a good chance of being implemented. In that the number and location of significant lesions, and the ACC-AHA 15,16 respect, models based on European data might more easily pene- lesion classification. An electronic case record form (eCRF) trate European practices than models based on US data. was used for data capture, which was programmed on the basis Furthermore, several risk models did not have separate training of the Cardiology Audit and Registration Data Standards (CARDS) for PCI.17,18 The eCRF was accessible via the Internet for data and validation cohorts.2,7,8,12 Without such separate cohorts, the entry and editing. Data were securely stored on a computer main- training data cannot not be formally validated. As a consequence, Downloaded from frame that was physically located in the European Heart House, their reliability remains uncertain. Nice, France. Automated edit checks were performed to search The Euro Heart Survey of PCIs (EHS-PCI) was developed to for missing data, contradictory data entries, as well as for values obtain quantitative information on the adherence to guidelines that are out of the specified normal range. Additionally, manual and outcomes in European patients undergoing PCI for different edit checks were performed by the data management staff of the

indications. The survey was undertaken during 2005–08 and European Heart House. Final editing of the data, as well as data ana- http://eurheartj.oxfordjournals.org/ includes data on 46 064 patients from European hospitals. Thus, lyses, was performed at the Institut fu¨r Herzinfarktforschung Lud- the EHS-PCI provides a unique opportunity to develop (and vali- wigshafen an der Universitaet Heidelberg (IHF), Ludwigshafen, date) a model for the prediction of patient prognosis after PCI, Germany. Any issues that appeared during this process were which reﬂects the modern clinical practice. In view of the large resolved in cooperation with the local investigators. The protocol of the EHS-PCI Registry was approved by each local number and large variety of hospitals that participated in Ethics Committee when required. All patients provided informed EHS-PCI, the results of this analysis will potentially be applicable consent for processing their data anonymously. to a broad variety of European practices.

Primary objective Methods by guest on April 2, 2016 The EHS-PCI Registry was designed to evaluate the application of PCI Registry within the Euro Heart Survey PCI-related treatment guidelines in routine clinical practice. With respect to patient outcome, the current study focuses on mortality. programme In this manner, endpoints that are vulnerable for observer bias, such The EHS programme of the European Society of Cardiology (ESC) as re-myocardial infarction (MI), are avoided and adjudication of such was originally designed as a series of surveys, to obtain information events is not required. All-cause mortality was reported by the local 13 on the application of clinical practice guidelines in the ESC investigators and not adjudicated by a Clinical Event Committee/ member countries, covering the broad spectrum of cardiology prac- Data Safety Monitoring Board. tice,14 an extensive descriptive paper about which is currently being composed. Typically, patient enrolment in EHSs was scheduled for short-term periods of 3–6 months, thus taking the risk of being Statistical analysis influenced by accidental, or just structural, season-bound variations When patient characteristics were incomplete, data were imputed. in patient management or events. In contrast, patient enrolment in Otherwise, the missing patient data might lead to biased estimates. the EHS-PCI Registry lasted for a period of 3 years, from May Since occasionally not all data can be collected in patients who die 2005 to April 2008. That period is long enough to level off acciden- early, patients with incomplete data often have a higher mortality. tal situations, as well as structural differences in patient management Missing values were imputed with the expected value according to between participating hospitals, which makes it greatly suitable for gender, age, and PCI indication. We also performed a sensitivity analy- our purpose. sis using multiple imputation methods. We found that all variables except prior renal failure (RF) and prior MI were of influence in all Patients and procedures models. These two variables were included in some models but not A total of 176 centres with PCI facilities from 33 ESC member in others (RF included in 6 of 20 models and prior MI included in 11 countries participated in the EHS-PCI Registry. The sample of of 20 models). As with regression imputation the best goodness of hospitals consisted of a mixture of tertiary referral university fit was achieved, this strategy was chosen. hospitals (48%), hospitals that could be considered satellites of uni- The percentage of missing data in variables that were significant in versity hospitals (15%), district or regional hospitals (13%), specialist multivariate analyses was: 9.1% for bifurcation lesion, 8.7% for haemo- cardiology centres (11%), community hospitals (8%), and private dynamic instability, 5.3% for valvular heart disease, 3.9% for TIMI flow, hospitals (5%). 3.4% for body mass index (BMI), and 3.2% for smoking status; other 1400 M. de Mulder et al. variables had fewer than 2% missing data. Discharge status, gender, age, procedural characteristics (which are listed in Table 1) and the incidence and PCI indication were known in all patients. of the primary objective. Variables that were associated with in-hospital The database was randomly divided into two equal parts by using death with a significance level of P , 0.5 entered the multivariate stage. A a specific application of the statistical analysis program. The first value of P , 0.5 was chosen in order not to miss any potential variables in part was used to develop the mortality risk score (‘training data the multivariate model. The final multivariable regression model was set’). The second part was used to validate the score (‘validation then constructed using the backward elimination of the least significant data set’). variables, until all variables had a significance level of P , 0.15. Sub- Univariate logistic regression analyses were applied on the training sequently, a mortality risk score was determined that included all vari- data set to study the association between a broad range of clinical and ables that composed the final regression model. The contribution of

Table 1 Baseline characteristics of the study patients

Training cohort Validation cohort P-value ...... Number of patients 23.032 23.032 Age (years) 64 (55, 72) 64 (56, 72) 0.85 Men 74 74 0.81

...... Downloaded from Indication for PCI Admission with STEMI 18 17 0.14 Admission with non-STEMI 13 13 0.73 Stabilized ACS 21 21 0.56 Elective procedure 49 49 0.37 ...... http://eurheartj.oxfordjournals.org/ Body mass index 27 (25, 30) 27 (25, 30) 0.63 Hypertension 69 70 0.53 Hypercholesterolaemia 64 65 0.21 Diabetes mellitus 25 25 0.81 Current smoker 27 27 0.25 Ever smoker 52 52 0.53 Prior PCI 24 25 0.36 Prior CABG 6.2 6.3 0.78

Prior myocardial infarction 34 34 0.64 by guest on April 2, 2016 Congestive heart failure 11 11 0.71 Peripheral vascular disease 6.0 6.0 0.83 Prior stroke 4.1 4.1 0.74 Chronic renal insufﬁciency 3.5 3.5 0.89 Valvular heart disease 2.1 2.3 0.22 ...... Number of diseased vessels 1 47 47 0.11 2 31 32 0.02 3 21 21 0.45 ...... Left main 4.5 4.6 0.53 Proximal LAD diseased 34 34 0.53 Bifurcation lesion 16 16 0.47 Type-C lesion 28 28 0.41 Haemodynamic instability (at presentation) 2.7 2.7 0.71 Transferred from other hospital 23 23 0.57 ...... Left ventricular functiona EF .50% 69 69 0.88 EF 41–50% 19 19 0.65 EF 31–40% 8.9 8.6 0.37 EF ,30% 4.1 4.0 0.79

Continuous data are presented as median values (25th–75th percentile); dichotomous data are presented as percentages. EF, ejection fraction. aBased on 32 267 patients EuroHeart score for the evaluation of in-hospital mortality after PCI 1401 these variables to the risk score was weighed according to the corre- Lemeshow (H–L) goodness-of-fit test] in the training and in the vali- sponding regression coefficient in the logistic model [i.e. the natural log- dation data set. arithm of the corresponding odds ratio (OR)]. All analyses were repeated for the cohort of patients who presented The performance of the mortality risk score was finally studied with with ST-elevation MI (STEMI). respect to discrimination (C-index) and calibration [Hosmer– The analyses were performed with SAS 9.1 software.

Table 2 Association between baseline characteristics and in-hospital mortality in the training cohort

In-hospital death Crude odds ratio and 95% CI Multivariable adjusted odds ratio 95% CI ...... Age, years (median) 71/64 1.05 (1.04–1.06) ...... Age categorized (years) ,50 0.73 1 — ≥50–60 0.83 1.1 (0.69–1.9) — ≥60–70 1.2 1.7 (1.03–2.7) 1.7 (1.2–2.5) ≥70–80 2.0 2.7 (1.7–4.3) 2.4 (1.7–3.4) ≥ 80 4.7 6.8 (4.2–11) 4.2 (2.8–6.5) Downloaded from ...... Female 2.2/1.2 1.8 (1.5–2.3) 1.6 (1.2–2.1) Body mass index ,25 2.2/1.1 2.0 (1.6–2.5) 1.8 (1.4–2.3) Hypertension 1.3/1.4 0.93 (0.74–1.2) — Hypercholesterolaemia 0.94/1.8 0.57 (0.46–0.71) — Diabetes mellitus 2.1/1.1 1.8 (1.4–2.2) 1.9 (1.5–2.5) http://eurheartj.oxfordjournals.org/ Ever smoker 1.2/1.3 0.94 (0.76–1.2) 1.4 (1.04–1.9) Prior PCI 0.77/1.6 0.52 (0.38–0.71) — Prior CABG 0.85/1.4 0.63 (0.36–1.09) 0.35 (0.18–0.69) Prior myocardial infarction 1.3/1.4 0.92 (0.73–1.2) — Congestive heart failure 1.5/1.4 1.1 (0.79–1.5) — Peripheral vascular disease 2.4/1.3 1.8 (1.3–2.6) — Prior stroke 3.2/1.3 2.4 (1.7–3.5) 1.8 (1.2–2.8) Chronic renal insufﬁciency 3.1/1.3 2.3 (1.5–3.5) —

Valvular heart disease 2.6/1.3 1.9 (1.1–3.4) 1.7 (0.83–3.4) by guest on April 2, 2016 ...... Number of diseased vessels 1 0.85 1 — 2 1.4 1.7 (1.3–2.2) — 3 2.9 3.5 (2.7–4.6) 1.4 (1.1–1.9) ...... Left main 5.3/1.3 4.2 (3.1–5.7) 2.2 (1.5–3.3) Proximal LAD diseased 2.4/1.0 2.4 (1.9–3.0) 1.6 (1.2–2.0) Bifurcation lesion 2.1/1.5 1.5 (1.1–1.9) 1.6 (1.1–2.1) Type-C lesion 2.6/1.0 2.6 (2.1–3.2) 1.5 (1.2–1.9) TIMI ﬂow 0/1 before PCI. 3.4/0.71 4.9 (3.9–6.2) 1.5 (1.2–2.1) ...... Indication for PCI Elective procedure 0.24 1 — Stabilized after ACS 0.86 3.6 (2.1–6.2) 2.6 (1.5–4.4) Admission with non-STEMI 2.1 9.1 (5.9–14) 5.0 (3.2–7.8) Admission with STEMI 5.4 24 (16–35) 7.8 (5.1–12) ...... Haemodynamic instability 29/0.83 52 (41–66) 17 (13–23) ...... Left ventricular functiona Class I (.50%) 0.46 1 — Class II (31–50%) 2.1 4.6 (3.2–6.4) — Class III (≤30%) 11.5 28 (19–40) —

Continuous data (age) are presented as median values; dichotomous data are presented as percentages. For in-hospital mortality, data represent mortality when variable is present (ﬁrst number) or absent (second number). Left ventricular function was not used for multivariate analysis, as in 30% of patients, this value was missing. 1402 M. de Mulder et al.

Results 3.1–5.7) were the most relevant angiographic characteristics for in-hospital death. Patient characteristics The EHS-PCI Registry enrolled a total of 46 064 PCI patients. The Mortality risk score median age of the study cohort was 64 years and 74% were men. A total of 16 variables remained in the multivariable model for the Fifty-one per cent of patients underwent PCI for (stabilized) acute prediction of in-hospital death (Table 2), among which haemo- coronary syndromes (ACS), and 49% had an elective procedure. In dynamic instability at admission, STEMI, and age ≥80 years were 94% of the patients, a stent was implanted; 46% of these stents most dominant. Ten variables were patient-related and could be were drug eluting. In 84% of the patients, percutaneous access obtained prior to the PCI procedure. Six factors were derived was via the femoral and in 15% via the radial approach. Patients during angiography. The multivariable model translated in the in the training and validation data sets had similar clinical and angio- scoring system is presented in Figure 1. There is a direct relation graphic characteristics (Table 1). Patients were discharged after 2 between the number of risk points and the estimated and days (inter-quartile range 1–4), 85.7% went home, 12.8% was observed mortality. For example, a 72-year-old (3 points) transferred to another hospital, and 1.5% to a rehabilitation centre. woman (2 points) with a prior stroke (2 points) but no known heart disease [thus no prior coronary artery bypass grafting Determinants of in-hospital mortality in (CABG), 4 points] who presents with STEMI (8 points) and left

the training data set main disease (3 points) has a total risk score of 22 points. The Downloaded from In the training cohort, a total of 339 patients (1.5%) died during hos- observed in-hospital mortality risk among the patients with 22 pitalization. In univariable analysis, advanced age, particularly age risk points was 5.3% (20 of 376 patients), and the predicted risk above 80 years [OR 6.8; 95% conﬁdence interval (CI) 4.2–11], (based on the model) was 3.8%. haemodynamic instability (i.e. cardiac shock at admission or resusci- As demonstrated in Figures 2 and 3, the majority of patients ≈ ≤ tation prior to PCI) (OR 52; 95% CI 41–66), left ventricular function ( 90%) have a low mortality risk, i.e. a score of 20 correspond- http://eurheartj.oxfordjournals.org/ (LVF) ≤30% (OR 28; 95% CI 19–40), and STEMI (OR 24; 95% CI ing with a mortality of ,2%. A score of 21–26 (2–8.4% mortality) 16–35) were strongly associated with increased mortality risk is present in 7.5% of the patients and the remaining 2.5% of the (Table 2). The presence of three-vessel disease (OR 3.5; 95% CI patients is a high-risk population with in-hospital mortality over 2.7–4.6) and the presence of left main disease (OR 4.2; 95% CI 7.5%, i.e. a score of ≥27. by guest on April 2, 2016

Figure 1 EuroHeart PCI score. Assigned integer scores. EuroHeart score for the evaluation of in-hospital mortality after PCI 1403

Model performance P-value 0.93). When subsequently the risk score is created and The multivariate training model has an excellent performance in applied to the training set, the C-index is 0.91 and H–L P-value terms of discrimination (C-index 0.91) and calibration (H–L 0.39. In the validation set, similar discrimination (C-index 0.90) and adequate calibration (H–L P-value 0.18) were observed (Figure 4). We also investigated the model performance in different subgroups (Table 3) and compared the performance of the current model with others (Table 4).

Patients presenting with ST-elevation myocardial infarction We performed a separate analysis of the 8060 patients who presented with ST elevation ACS to have a valid model for this high- risk population, since only a small proportion of the original data consist of high-risk patients. From the original training and validation cohorts, the STEMI patients were selected, i.e. 4091 and 3969, respectively. In the training cohort, a total of 220 out of 4091 patients (5.4%) and in the vali- Downloaded from dation cohort 203 out of 3969 (5.1%) died during hospitalization. With multivariate analysis, 19 variables remained of signiﬁcant inﬂuence. Particularly, haemodynamic instability at admission (OR 14; 95% CI 10–20), age ≥80 (OR 4.6; 95% CI 2.7–7.9), and left main disease (OR 2.1; 95% CI 1.2–3.7) were associated with a http://eurheartj.oxfordjournals.org/ Figure 2 Distribution of assigned scores over the validation high in-hospital mortality risk (Table 5). cohort. In this subpopulation, the area under the receiver-operating characteristic (ROC) curve was 0.86 with an H–L P-value of by guest on April 2, 2016

Figure 3 The EuroHeart PCI score model and observed in-hospital mortality. (Top left) Score in the training cohort. (Top right) Score in the validation cohort. (Bottom) Enlargement of patients with an intermediate score (10–30 points) in the training (left) and validation cohort (right) to better illustrate the transition point from low to intermediate risk. Score .40 is not illustrated as these score groups contain few (≤6) patients, which makes accurate prediction more difﬁcult. Data points correspond to the observed mortality (y-axis) for patients with a particular score (x-axis), and line represents predicted mortality. 1404 M. de Mulder et al. Downloaded from http://eurheartj.oxfordjournals.org/ by guest on April 2, 2016 Figure 4 Expected vs. observed in-hospital mortality. (Top left) Training cohort. (Top right) Validation cohort. (Bottom) Enlargement of both cohorts. Rates were calculated with the Hosmer–Lemeshow goodness-of-ﬁt test.

0.42, indicating a good discriminatory value in this high-risk subpopulation. Subsequently, we again created a simplified scoring Table 3 Subgroup validation in validation cohort model which was then tested in the training and validation data Validated subgroup Sample/mortality (n) C-index H–L sets, the score ranged from 2 to 37 points (Figure 5). P-value In the training STEMI data, this simplified model demonstrated ...... an area under the ROC curve of 0.86 with an H–L P-value of Male 17 112/180 0.90 0.008 0.75. In the STEMI validation data, similar discrimination (C-index Female 5920/125 0.90 0.94 0.89) and calibration (H–L P-value 0.70) were observed. This Age ≥70 years 7433/191 0.88 0.36 acknowledges the validity of the separate STEMI model. Age ,70 years 15 599/114 0.89 0.003 Diabetes 5772/119 0.90 0.80 No diabetes 17 084/192 0.91 0.67 Discussion Patient in shock 574/157 0.74 0.85 Patient not in shock 22 458/148 0.82 0.21 We developed a risk score for in-hospital mortality after PCI based PCI in ACS patient 11 710/279 0.91 0.85 on clinical and angiographic data from the EHS-PCI database, with a PCI in elective patient 11 291/25 0.57 0.81 high discriminatory value, and demonstrated its value in contem- STEMI 3969/203 0.89 0.57 porary practice. Strong points of our model are the large sample No STEMI 19 063/102 0.81 0.37 size, pan-European multicentre approach, and the use of recent data from everyday clinical practice. Although the H–L P-value is significant in men and patients ,70, the maximum Previous work in patients undergoing CABG resulted in the difference in observed and expected mortality per tentile is only four deaths; this was observed in the high-risk group (tentile 9/10 for both subgroups). H–L, EUROSCORE, a tool designed to assess the peri-operative risk Hosmer–Lemeshow; STEMI, ST-elevation myocardial infarction. for heart surgery.19 Recently, this model was also tested in PCI EuroHeart score for the evaluation of in-hospital mortality after PCI 1405

Table 4 Different risk models for percutaneous coronary intervention outcomes

Author and year of AUC Predicted Multi DES Remarks publication (C-index) endpoint centre used? ...... This study 0.91 In-hospital Yes Yes Based on European population, contemporary mortality practice Peterson et al.10 (2010) 0.93 In-hospital Yes Yes Based on North American population, contemporary mortality practice Singh et al.8 (2008) 0.78 In-hospital MACE No N/A Expansion of MCRS model with CAD-speciﬁc index 0.75 In-hospital mortality Madan et al.3 (2008) 0.70 MACE at 30 days No Yes Adding morbidity may have lowered discriminatory ability Negassa et al.4 (2007) 0.82 In-hospital Yes N/A 3 factors in risk model mortality Halkin et al.2 (2005) 0.83 30-day mortality Yes No Patients in shock or with complex coronary anatomy 0.79 1-year mortality were excluded Addala et al.9 (2004) 0.78 6-month mortality Yes No STEMI patients from various PAMI trials Downloaded from Qureshi et al.5 (2003) 0.87 In-hospital No No LVF and lesion characteristics not included mortality Shaw et al.7 (2002) 0.89 In-hospital Yes No No systematic data auditing across participating mortality centres Large data set (.100 000 PCIs) 6 Moscucci et al. (2001) 0.90 In-hospital Yes No Little high-risk procedures http://eurheartj.oxfordjournals.org/ mortality Designed as bedside tool with only clinical parameters Rihal et al.12 (2000) 0.86 Death after PCIa No No 45% of procedures only balloon angioplasty

AUC, area under the ROC curve; DES, drug-eluting stent; MACE, major adverse cardiac event; N/A, not available; MCRS, Mayo Clinic Risk Score; CAD-specific index was developed to determine prognostic influence of co-morbid conditions. aNo specific time frame specified.

Table 5 Association between baseline characteristics and in-hospital mortality in the ST elevation acute coronary by guest on April 2, 2016 syndrome training cohort

In-hospital deatha (%) Crude odds ratio and 95% CI Multivariable adjusted odds ratio 95% CI ...... Age (years) 71/62 1.04 (1.03–1.06) — ...... Age categorized (years) ,50 2.4 1 — ≥50–60 3.1 1.3 (0.72–2.3) — ≥60–70 5.4 2.3 (1.3–3.9) 1.9 (1.2–3.0) ≥70–80 7.9 3.4 (2.0–5.8) 2.8 (1.8–4.3) ≥80 11.9 5.4 (3.1–9.6) 4.6 (2.7–7.9) ...... Female 7.6/4.6 1.7 (1.3–2.3) 1.5 (1.06–2.2) Body mass index 27/26 0.97 (0.94–1.01) — Body mass index ,25 7.4/4.2 1.8 (1.3–2.4) 1.8 (1.3–2.5) Hypertension 5.4/3.9 1.4 (1.04–1.9) — Hypercholesterolaemia 4.2/4.8 0.89 (0.67–1.2) — Diabetes mellitus 7.9/4.1 1.9 (1.4–2.6) 1.7 (1.2–2.4) Ever smoker 4.3/4.6 0.88 (0.66–1.2) 1.8 (1.2–2.7) Prior PCI 4.8/4.9 0.98 (0.62–1.6) — Prior CABG 4.9/5.2 1.06 (0.40–2.8) 0.30 (0.10–0.94) Prior myocardial infarction 7.7/4.4 1.8 (1.3–2.4) — Continued 1406 M. de Mulder et al.

Table 5 Continued

In-hospital deatha (%) Crude odds ratio and 95% CI Multivariable adjusted odds ratio 95% CI ...... Congestive heart failure 8.8/4.7 1.9 (1.1–3.1) — Peripheral vascular disease 12.7/4.6 2.8 (1.7–4.6) — Prior stroke 11.3/4.7 2.5 (1.5–4.2) 1.6 (0.85–3.0) Chronic renal insufﬁciency 10.7/4.8 2.3 (1.1–4.5) — Valvular heart disease 12.5/4.8 2.7 (1.07–6.9) — ...... Number of diseased vessels 1 3.3 1 — 2 5.6 1.7 (1.2–2.5) 1.4 (0.92–2.0) 3 9.1 2.9 (2.1–4.1) 1.5 (1.01–2.3) ...... Left main 20.4/4.8 5.1 (3.4–7.7) 2.1 (1.2–3.7) Proximal LAD diseased 8.0/3.9 2.1 (1.6–2.8) 1.4 (1.01–2.0) LAD is target vessel 6.3/4.7 1.4 (1.04–1.8) 1.4 (0.96–1.9) Downloaded from Bifurcation lesion 8.7/5.3 1.7 (1.2–2.5) 1.4 (0.93–2.2) Type-C lesion 8.1/3.9 2.2 (1.6–2.9) 1.5 (1.08–2.0) ...... PCI indication Primary PCI (,24 hrs) 5.3 1 —

Rescue PCI 7.0 1.35 (0.83–2.2) — http://eurheartj.oxfordjournals.org/ Facilitated PCI 4.1 0.75 (0.33–1.7) — ...... TIMI ﬂow 0/1 before PCI 6.3/3.5 1.9 (1.3–2.6) 1.5 (1.05–2.3) Haemodynamic instability 30.4/2.7 17 (13–23) 14 (10–20) ...... Left ventricular functionb Class I (.50%) 1.9 1 — Class II (31–50%) 4.0 2.2 (1.3–3.6) — Class III (≤30%) 26.8 19 (11–32) — ...... Ischaemic time (h) by guest on April 2, 2016 0–3 4.7 1 — 3–6 4.8 1.03 (0.71–1.5) — 6–12 5.1 1.09 (0.71–1.7) — .12 6.4 1.4 (0.90–2.1) 1.4 (0.92–2.1)

Continuous data (age, BMI) are presented as median values; dichotomous data are presented as percentages. Qualitative estimated based on 4091 patients. aFor in-hospital mortality, data represent mortality when variable is present (ﬁrst number) or absent (second number). bLVF was not included in the multivariable analyses as 36% of patients had missing data.

and demonstrated a good discriminatory value.20 However, the P-value of 0.05. Thus, discrimination is good, but calibration is EUROSCORE includes various operation-related factors that per poor. definition do not apply to PCI. Thus, for PCI patients, not all Our model overcomes these limitations and may therefore be a items that compose the score can be filled, which, again per defi- good first step to create a specific European risk score to assess nition, results in inappropriate risk estimation. the peri-procedural risk of PCI. Next, it might be used as a bench- These limitations were overcome by the recent NCDR model.10 mark tool to compare different hospitals. However, additional To avoid unnecessary risk models, we tried to validate this model testing in a separate clinical cohort with new data may be con- on our data; there are some limitations however. The focus in data sidered beforehand. collection in the EHS-PCI survey was different from the NCDR Subgroup analysis revealed that our model is less useful to data. As a result, we did not have information on three (out of predict mortality in patients who undergo an elective procedure, eight) variables, i.e. glomerular filtration rate, New York Heart i.e. not ACS-related. Apparently, it is difficult to predict events in Association class, and chronic lung disease. Furthermore, the this group as the mortality risk is very low (25 events in 11 291 NCDR classification of the indication for PCI was different. With patients ¼ 0.22%). Perhaps, we have to accept that mortality risk these limitations, we found a c-statistic of 0.89 with an H–L in elective procedures cannot be predicted with classical risk EuroHeart score for the evaluation of in-hospital mortality after PCI 1407 Downloaded from http://eurheartj.oxfordjournals.org/

Figure 5 EuroHeart STEMI PCI score. Assigned integer scores. Score until 31 points; Scores 32–37 are left out as these contained only few patients (≤13). by guest on April 2, 2016 factors, but might be more dependent on other factors such as on routine procedures. Since we realize that registry designs are operator experience or (contrast) allergies. susceptible to observer bias, especially with regard to ‘soft’ par- Another point of interest is that the tentile of patients with the ameters, we chose the incidence of all-cause mortality during hos- highest risk has a mortality of 10%. Perhaps that further stratifi- pitalization as the primary endpoint of this study. Particularly, this is cation of this subcohort may improve the calibration. the clinically most relevant endpoint for patients. Major determinants in our risk model are haemodynamic The predictive value of future risk models might be further instability, STEMI, age ≥80, and three-vessel disease. Singh enhanced when they are also fitted with serum markers such as et al.21 gave an overview of variables used in different risk admission glucose,22 C-reactive protein,23 and N-terminal-pro-brain scores. The main factors they described are also included in our natriuretic peptide.24 model. However, additional factors from their analysis such as It is important to recognize that interventional cardiology is RF and peripheral artery disease did not contribute significantly under continuous development and new techniques arise which in our multivariate analysis. Possibly, this is a consequence of the will require adaptation of existing predictive models. However, it limitations of a survey in which these data might not have been col- might be sufficient to re-validate a powerful existing model lected as precisely as in a clinical trial. Indeed, non-collection of instead of developing a complete new one. variables is a problem for any risk model, particularly when it is externally validated. Limitations It appeared that prior CABG has a protective effect in our Our analysis has several limitations that should be mentioned. score model. Interestingly, out of 2857 patients with prior Since our model predicts in-hospital mortality, this can be influ- CABG, 2042 (72%) had the intervention only in their native enced with different discharge policies. For example, referral vessels. Therefore, we might speculate that these interventions centres where patients quickly after PCI are transported to a were done under the protection of patent bypass grafts, resulting nearby hospital for further recovery may have low mortality in better outcomes. figures as patients spend only several hours in that particular refer- The protocol did not mandate serial electrocardiograms or ral centre (this was applicable to 12.8% of patients). The same blood sampling for the determination of cardiac enzymes. As per might be relevant for centres without on-site surgical backup design, it was the intention to minimize the impact of the protocol when emergency CABG as a result of the PCI is required. 1408 M. de Mulder et al.

However, only in 49 patients (0.1%), emergency CABG was 7. Shaw RE, Anderson HV, Brindis RG, Krone RJ, Klein LW, McKay CR, Block PC, performed. Shaw LJ, Hewitt K, Weintraub WS. Development of a risk adjustment mortality model using the American College of Cardiology-National Cardiovascular Data Another matter is selection bias as to who receives angiography. Registry (ACC-NCDR) experience: 1998–2000. J Am Coll Cardiol 2002;39: It is conceivable that clinicians decide not to perform angiography 1104–1112. as they consider that as a result of the advanced age, particularly 8. Singh M, Rihal CS, Roger VL, Lennon RJ, Spertus J, Jahangir A, Holmes DR Jr. Comorbid conditions and outcomes after percutaneous coronary intervention. over 80, PCI risk is already too high. Knowledge of the coronary Heart 2008;94:1424–1428. anatomy would not change their therapy. Consequently, PCI risk 9. Addala S, Grines CL, Dixon SR, Stone GW, Boura JA, Ochoa AB, Pellizzon G, in frail octogenarians may actually be underestimated. O’Neill WW, Kahn JK. Predicting mortality in patients with ST-elevation myocardial infarction treated with primary percutaneous coronary intervention (PAMI Our model uses only patient-related factors; therefore, we are risk score). Am J Cardiol 2004;93:629–632. not informed on operator experience and procedure volumes, 10. Peterson ED, Dai D, DeLong ER, Brennan JM, Singh M, Rao SV, Shaw RE, Roe MT, which also affect outcomes. In small hospitals, for example, experi- Ho KKL, Klein LW, Krone RJ, Weintraub WS, Brindis RG, Rumsfeld JS, Spertus JA, on behalf of the NCDR Registry Participants. Contemporary mortality risk pre- ence may be lower due to small sample size. When these are taken diction for percutaneous coronary intervention: results from 588,398 procedures into account, the predictive value (as expressed by the C-index) of in the National Cardiovascular Data Registry. J Am Coll Cardiol 2010;55: future models might further increase. However, then the model is 1923–1932. 11. Matheny ME, Ohno-Machado L, Resnic FS. Discrimination and calibration of mor- not suitable for benchmarking purposes. Additional parameters, tality risk prediction models in interventional cardiology. J Biomed Inform 2005;38: such as heart rate or novel biomarkers, might give an even 367–375. better discriminatory power, but were regrettably not available. 12. Rihal CS, Grill DE, Bell MR, Berger PB, Garratt KN, Holmes DR Jr. Prediction of death after percutaneous coronary interventional procedures. Am Heart J 2000;

139:1032–1038. Downloaded from 13. Silber S, Albertsson P, Aviles FF, Camici PG, Colombo A, Hamm C, Jorgensen E, Conclusion Marco J, Nordrehaug JE, Ruzyllo W, Urban P, Stone GW, Wijns W. Guidelines for percutaneous coronary interventions. The Task Force for Percutaneous Coron- We have developed a risk score to predict in-hospital mortality in ary Interventions of the European Society of Cardiology. Eur Heart J 2005;26: PCI patients. This model is based on a large patient sample, 804–847. 14. Gitt AK, Zeymer U, Zahn R, Weidinger F, Seabra-Gomes R, Eberli F, Serruys P, http://eurheartj.oxfordjournals.org/ pan-European multicentre approach, and recent data from every- Bauer T, Hamm C, Marco J. Current practice of PCI for ACS and stable angina day clinical practice, which strengthens its applicability. As it uses in Europe 2005–2008: lessons from the Euro Heart Survey PCI registry. Eur clinical and angiographic data, it is easy to implement in clinical Heart J 2009;30 (Abstract Supplement): 859. 15. Ryan TJ, Faxon DP, Gunnar RM, Kennedy JW, King SB III, Loop FD, Peterson KL, practice to estimate the in-hospital mortality risk after PCI. Reeves TJ, Williams DO, Winters WL Jr. Guidelines for percutaneous transluminal coronary angioplasty. A report of the American College of Cardiology/Amer- ican Heart Association Task Force on Assessment of Diagnostic and Therapeutic Funding Cardiovascular Procedures (Subcommittee on Percutaneous Transluminal Cor- We acknowledge the following companies who have supported the onary Angioplasty). Circulation 1988;78:486–502. PCI-Registry: main sponsors—Boston Scientiﬁc, GlaxoSmithKline; 16. Ellis SG, Vandormael MG, Cowley MJ, DiSciascio G, Deligonul U, Topol EJ, sponsors—Bristol-Myers Squibb, Eli Lilly. Bulle TM. Coronary morphologic and clinical determinants of procedural outcome with angioplasty for multivessel coronary disease. Implications for

patient selection. Multivessel Angioplasty Prognosis Study Group. Circulation by guest on April 2, 2016 Conflict of interest: C.H. reports a Lecture honorarium and/or 1990;82:1193–1202. advisory board position from Medtronic, Abbott, Boston Scientific, 17. http://www.escardio.org/Policy/Documents/CARDS-dataset-PCI-1104.pdf (11 and Braun. November 2010). 18. Flynn MR, Barrett C, Cosio FG, Gitt AK, Wallentin L, Kearney P, Lonergan M, Shelley E, Simoons ML. The Cardiology Audit and Registration Data Standards References (CARDS), European data standards for clinical cardiology practice. Eur Heart J 1. Srinivas VS, Hailpern SM, Koss E, Monrad ES, Alderman MH. Effect of physician 2005;26:308–313. volume on the relationship between hospital volume and mortality during 19. Roques F, Michel P, Goldstone AR, Nashef SA. The logistic EuroSCORE. Eur primary angioplasty. J Am Coll Cardiol 2009;53:574–579. Heart J 2003;24:881–882. 2. Halkin A, Singh M, Nikolsky E, Grines CL, Tcheng JE, Garcia E, Cox DA, Turco M, 20. Romagnoli E, Burzotta F, Trani C, Siviglia M, Biondi-Zoccai GG, Niccoli G, Stuckey TD, Na Y, Lansky AJ, Gersh BJ, O’Neill WW, Mehran R, Stone GW. Leone AM, Porto I, Mazzari MA, Mongiardo R, Rebuzzi AG, Schiavoni G, Prediction of mortality after primary percutaneous coronary intervention for Crea F. EuroSCORE as predictor of in-hospital mortality after percutaneous cor- acute myocardial infarction: the CADILLAC risk score. J Am Coll Cardiol 2005; onary intervention. Heart 2009;95:43–48. 45:1397–1405. 21. Singh M, Rihal CS, Lennon RJ, Garratt KN, Holmes DR Jr. A critical appraisal of 3. Madan P, Elayda MA, Lee VV, Wilson JM. Predicting major adverse cardiac events current models of risk stratification for percutaneous coronary interventions. after percutaneous coronary intervention: the Texas Heart Institute risk score. Am Heart J 2005;149:753–760. Am Heart J 2008;155:1068–1074. 22. de Mulder M, Cornel JH, van der Ploeg T, Boersma E, Umans VA. Elevated admis- 4. Negassa A, Monrad ES, Bang JY, Srinivas VS. Tree-structured risk stratification of sion glucose is associated with increased long-term mortality in myocardial infarc- in-hospital mortality after percutaneous coronary intervention for acute myocar- tion patients, irrespective of the initially applied reperfusion strategy. Am Heart J dial infarction: a report from the New York State percutaneous coronary inter- 2010;160:412–419. vention database. Am Heart J 2007;154:322–329. 23. Iijima R, Byrne RA, Ndrepepa G, Braun S, Mehilli J, Berger PB, Schomig A, 5. Qureshi MA, Safian RD, Grines CL, Goldstein JA, Westveer DC, Glazier S, Kastrati A. Pre-procedural C-reactive protein levels and clinical outcomes after Balasubramanian M, O’Neill WW. Simplified scoring system for predicting mor- percutaneous coronary interventions with and without abciximab: pooled analysis tality after percutaneous coronary intervention. J Am Coll Cardiol 2003;42: of four ISAR trials. Heart 2009;95:107–112. 1890–1895. 24. de Winter RJ, Stroobants A, Koch KT, Bax M, Schotborgh CE, Mulder KJ, 6. Moscucci M, Kline-Rogers E, Share D, O’Donnell M, Maxwell-Eward A, Sanders GT, van Straalen JP, Fischer J, Tijssen JG, Piek JJ. Plasma N-terminal Meengs WL, Kraft P, DeFranco AC, Chambers JL, Patel K, McGinnity JG, pro-B-type natriuretic peptide for prediction of death or nonfatal myocardial Eagle KA. Simple bedside additive tool for prediction of in-hospital mortality infarction following percutaneous coronary intervention. Am J Cardiol 2004;94: after percutaneous coronary interventions. Circulation 2001;104:263–268. 1481–1485.